HP preps Project Kraken for monster HANA in-memory jobs
Sixteen Ivy Bridge-EX sockets and 12TB in a single image
HP has revealed a little more about its "Project Kraken" in-memory system that it is cooking up in conjunction with the engineers at SAP. It's talking about a future in which there are lots of scale-out servers like its Project Moonshot systems and big-memory systems like Kraken on the other end of the spectrum – with not as much plain-vanilla, general-purpose iron in between.
While server makers are building clusters to support the HANA in-memory database, which has been able to back-end SAP's Business Suite application software since the beginning of the year, Paul Miller, vice president of converged systems at HP, tells El Reg that some customers are going to need larger single-system images to run their in-memory applications, much larger than can be built today with two-socket or four-socket Xeon or Opteron servers.
And thus, one of the first fruits of HP's "Project Odyssey" server development effort to take technologies from the Itanium-based Integrity and Superdome lines and recast them with Xeon-based iron, will be the Project Kraken system that was previewed at SAP's Sapphire Now user and partner event in Orlando, Florida on Thursday.
HP is being a little sketchy on the details about the Kraken server, named after the mythical beasts from the depths of the sea in the North Atlantic, but Miller gave El Reg a few details to whet the appetite.
The server will span up to sixteen processor sockets and will be based on Intel's future "Ivy Bridge-EX" Xeon E7 processors. The Xeon E7s are currently stuck at the "Westmere-EX" level, and machines using the E7-4800 and E7-8800 processors are the only ones that SAP has certified to run its HANA in-memory database. (Actually, if you want to be super-precise, only the 2.4GHz versions of those two families of chips are supported. SAP wants to make sure all HANA appliance makers use the exact same motors.)
Intel did not launch a "Sandy Bridge-EX" variant of the Xeon E7s, but is expected to get the Ivy Bridge-EX variants of the chip out the door in the fourth quarter.
Miller also told El Reg that the Kraken machine would have up to 12TB of main memory, addressable by all of the sockets. This is a much larger memory footprint than most relational databases have today, says Miller, adding that it is difficult to find a single database with 6TB, 9TB, or 12TB of data.
The HANA database has compression features that crunch the data down by a 2:1 ratio, so a 12TB memory footprint of the Kraken machine will be able to house a production database that would otherwise take up 24TB of memory and disk space on a machine that was not running the database in memory. At the moment, HP and SAP engineers are prototyping a system that scales up to 6TB of main memory in a single image.
Currently, HANA appliance makers are tending to build their machines based on a four-socket Xeon E7 v1 machine with 512GB per node, as prescribed by the very stringent rules SAP has created for HANA appliances. (All appliances run SUSE Linux Enterprise Server 11, too, which simplifies tuning and support for SAP.)
To scale up processing capacity, you lash nodes together and scale up as high as 8TB across a cluster of machines. But the HANA database on that cluster is not a single image. Each node has a copy of HANA and it runs multiple database tables that sit behind applications, just as happens in Business Suite production environments today on relational databases. Companies have hundreds of thousands of databases, not just one.
The Kraken server is aimed at supporting very large databases in memory, and this is something that HP is betting will be a big deal in the years ahead – as big of a deal as having dozens of microservers sporting x86 and ARM processors in a single chassis for supporting infrastructure and application workloads.
"We believe that a big portion of the market is either going to go to these hyperscale systems that scale out, but then also another big portion of the market is going to go to in-memory computing," explains Miller.
"When you can start to put your databases and applications in memory and not have to translate them from system to system," he said, "the programming model becomes simpler, the scalability becomes massive. If you look three to five years out, we believe that a lot of work will be based on Moonshot and a lot will be based on in-memory, and the kind of general purpose server is going to be less of a factor in the marketplace. That is why we are betting high on the bookends. We just see such an economic advantage on both ends."
HP is not saying when it expects to get the Kraken server tailored for HANA databases to market, but it probably stands to reason that Chipzilla is trying to launch the Ivy Bridge-EX sometime around Intel Developer Forum in San Francisco this September 10 through 13. Depending on how the Xeon E7 v3 (as it will presumably be called to keep in synch with other Ivy Bridge chips) ramps up, that could mean HP can get Kraken systems in the field in late 2013 or early 2014.
HP has not divulged how it will forge the sixteen-socket Odyssey box that will be the basis of the Kraken in-memory server, but it stands to reason that it will take its homegrown sx3000 chipset and Superdome 2 crossbar fabric and adapt it to the Xeon E7 v3 chips.
It would have been nice if Intel would have put the Xeon E7 chips and the "Poulson" Itanium 9500 and future "Kittson" Itanium chips into the same socket, as has been the on-again, off-again plan for years. But Intel killed off the Itanium-Xeon E7 convergence plan this February.
In addition to the Kraken preview, HP said that its AppSystem HANA appliances, which come with 1TB, 2TB, and 4TB memory configurations across clusters of ProLiant DL980 four-socket Xeon E7 v1 servers and which have been supporting HANA databases since they were available, have now been grandfathered in to run Business Suite apps on top of HANA. Miller says HP has hundreds of customers using HANA on the AppSystem appliances, by the way.
And because moving a database is a big deal for most customers, HP's Enterprise Services division has fired up tools to help move relational databases into HANA, and is also reminding everyone that it has hosted versions of Business Suite, with or without HANA underneath it, available as well. ®